| According to some predictions, the computing industry will have a programming expenditure of ;Eight classes of software error cause are defined and errors encountered in the experiment are classified according to these classes. It was not surprising to find that specification errors were the most serious and most difficult to detect and most prevalent in the English language specification.;The increase in reliability seen in multi-version software over individual-version software was substantial; it was even possible to combine three faulty versions and produce a combination that was completely fault-tolerant. This study shows the feasibility of multi-version software as an alternative to fault avoidance, at least in a research environment; the results presented here are sufficiently encouraging to warrant further and more intensive efforts, and, to this end, the next generation of experiments is defined.;During early 1981 in the UCLA Computer Science Department, 32 programmers were hired to take part in the experiment. Formal and informal specification methodologies were compared with 12 of the programmers writing software from a specification written in the formal specification language OBJ, 10 using the informal specification language PDL, and 10 using an English language specification. It was found that formal specification languages, while showing promise for the future, are presently very difficult to use and understand, and are severely limited in power. Based on our experimental experience, we present suggestions for improvements to formal specification techniques. |
